The future chooses a main body.
A brief Φ-forward push did not make every future look the same. It made different futures agree more about how much of the organism would consolidate into one dominant component.
Push once. Then let eight futures answer.
We returned to the same 38 early developmental states used in the dose experiment. Each state received the already-selected strong tangent intervention—or a control—and was then released into eight new random futures. No founder actions were scripted after the push.
Φ-forward canalized dominant-body allocation.
Across eight futures, we measured how much the fraction of mass in the largest connected component varied from future to future. Smaller spread means the futures agree more strongly about the dominance of the main body.
More dominant
Mean largest-component fraction, Φ-forward versus Φ-backward, passages 5–12. Difference +0.0173; 95% interval [+0.0064, +0.0307].
More consistent
Future-to-future breadth, Φ-forward versus Φ-backward. Difference −0.0136; 95% interval [−0.0251, −0.00181].
| Comparison | Breadth change | 95% interval | Families |
|---|---|---|---|
| Φ-forward − Φ-backward | −0.01358 | [−0.02509, −0.00181] | 14/19 narrower |
| Φ-forward − untouched | −0.02000 | [−0.03479, −0.00640] | 14/19 narrower |
| Φ-forward − blind swap | −0.01640 | [−0.02848, −0.00471] | 15/19 narrower |
| Φ-backward − untouched | −0.00643 | [−0.01986, +0.00648] | direction unresolved |
Not an immediate scar. A late developmental convergence.
For the first several passages, forward and backward futures overlap. The separation grows after passage 5 and becomes clearest at passages 10–12. During that late window, Φ-forward reduced breadth by 0.0199 versus Φ-backward and by 0.0272 versus untouched.
It did not flatten the whole space of forms.
The same intervention left the fine 32×32 morphology repertoire, founder-composition repertoire, and component-count breadth broadly unchanged. What tightened was a particular global decision: how much mass ends up in the dominant connected body.
Fine morphology stays open
Anchored repertoire difference −0.000081, interval [−0.00320, +0.00299]. The futures still explore distinct detailed shapes.
Identity mixture stays open
Founder-composition repertoire difference −0.00197, interval [−0.00647, +0.00210]. This is not simple founder homogenization.
Fragment count stays open
Component-count breadth difference +0.0113, interval [−0.535, +0.574]. The effect concerns dominance, not merely the number of pieces.
The physical-action control collapses exactly.
Four states had zero directional leverage: Φ-forward and Φ-backward selected the same physical swap. Across every passage and every repertoire readout, their maximum difference was exactly zero.
Every leave-one-family-out estimate of the main breadth effect remained negative: [−0.01673, −0.01086]. The median family effect (−0.01325) and trimmed mean (−0.01380) nearly equal the full estimate.
Information geometry can steer a developmental invariant.
The earlier dose experiment showed that the tangent can retain or erase causal history. This experiment goes one step closer to organism emergence: the same early information direction changes whether later environments produce a reliably dominant main body.
Not “one inevitable shape”
The fine-form repertoire stays open. Canalization happens on a coarser organizational axis.
Not “just more integration”
The intervention is brief and local. Its signature arrives later as agreement across independently randomized futures.
A candidate commitment variable
Dominant-body allocation may be one of the macroscopic decisions that information geometry helps stabilize before visible form is fully settled.
Find the commitment aperture.
We now have a causal readout that is both delayed and organism-level. The clean next experiment is to apply the same fixed strong tangent before, during, and after the developmental turn.
Timing intervention
At what moment does Φ-forward gain—and then lose—the ability to canalize a dominant main body across future environments?
What was actually measured.
The intervention candidates were inherited unchanged from the prior outcome-blind tangent selection. All four treatments shared the same eight new partition and reservoir seeds. Repertoire width was computed from all 28 future pairs. Family uncertainty resampled the 19 family means, with both schedules retained inside each family.